Re: kmemleak in apei_res_add

[Catalin Marinas <catalin.marinas@arm.com>]

On 12 August 2013 03:45, Alexandra N. Kossovsky
<Alexandra.Kossovsky@oktetlabs.ru> wrote:
> When running 3.10.3 with kmemleak enabled, I see following warnings from
> kmemleak:
>
> unreferenced object 0xffff88024d2219a0 (size 32):
>   comm "swapper/0", pid 1, jiffies 4294894532 (age 26865.180s)
>   hex dump (first 32 bytes):
>     58 de 0c 73 02 88 ff ff 58 de 0c 73 02 88 ff ff  X..s....X..s....
>     18 50 fd 7d 00 00 00 00 3f 50 fd 7d 00 00 00 00  .P.}....?P.}....
>   backtrace:
>     [<ffffffff813ddf40>] kmemleak_alloc+0x21/0x3e
>     [<ffffffff8111cb99>] kmemleak_alloc_recursive.constprop.58+0x16/0x18
>     [<ffffffff8111efac>] kmem_cache_alloc_trace+0xe6/0x12f
>     [<ffffffff81278f0e>] apei_res_add+0xb9/0xe8
>     [<ffffffff81278faa>] collect_res_callback+0x6d/0x7c
>     [<ffffffff812788ed>] apei_exec_for_each_entry+0x78/0x93
>     [<ffffffff8127891d>] apei_exec_collect_resources+0x15/0x17
>     [<ffffffff81b010e5>] erst_init+0x278/0x2a3
>     [<ffffffff810020b5>] do_one_initcall+0x7b/0x116
>     [<ffffffff81ac9ee7>] kernel_init_freeable+0x13f/0x1cc
>     [<ffffffff813db19f>] kernel_init+0x9/0xd6
>     [<ffffffff813fe5bc>] ret_from_fork+0x7c/0xb0
>     [<ffffffffffffffff>] 0xffffffffffffffff

Are there any "APEI:" errors in the kernel log? It could be some
freeing missing on an error path.

apei_res_add() allocates struct apei_res objects and places them on
the erst_resources.{iomem,ioport} list heads as given by erst_init().
The erst_resources is just a local variable on the erst_init() stack,
hence kmemleak doesn't have any trace of the allocated objects.

It looks like erst_init() calls apei_resources_request() which should
merge the local erst_resources list heads with the global
apei_resources_all but for whatever reason, this or any of the
previous calls in erst_init() fail and we don't get the merging. Some
of the goto err_fini in this function should probably be err_release.
I cc'ed Huang as the driver author.

-- 
Catalin